#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: Type: bug | Status: new Priority: high | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Changes (by RyanGlScott): * milestone: => 8.0.1 Comment: I'm going to set the milestone to 8.0.1, since this affects the upcoming GHC 8.0 and I'd hate to see so much `Typeable`-related code break. Please change the milestone if you disagree. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:2 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: Type: bug | Status: new Priority: high | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Changes (by RyanGlScott): * differential: => Phab:D1757 Comment: I still haven't figured out what's going on with `typeOf (Proxy :: Proxy 'Compose)`, but we can at least be comforted by the fact that executing it wasn't possible before GHC 8.0, since `Compose` is poly-kinded. The "regression" here is that `*` now shows up as `TYPE 'Lifted`, which I don't think is desirable. I've opened Phab:D1757 to fix that. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:4 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: Type: bug | Status: new Priority: high | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Comment (by Ben Gamari ): In [changeset:"65b810bdda625f2e98069c2c56ec93e1c65667a6/ghc" 65b810b/ghc]: {{{ #!CommitTicketReference repository="ghc" revision="65b810bdda625f2e98069c2c56ec93e1c65667a6" Show TYPE 'Lifted/TYPE 'Unlifted as */# in Show TypeRep instance Kind equalities changed how `*`/`#` are represented internally, which means that showing a `TypeRep` that contains either of those kinds produces a rather gross-looking result, e.g., ```

typeOf (Proxy :: Proxy 'Just) Proxy (TYPE 'Lifted -> Maybe (TYPE 'Lifted)) 'Just

We can at least special-case the `Show` instance for `TypeRep` so that
it prints `*` to represent `TYPE 'Lifted` and `#` to represent `TYPE
'Unlifted`.

Addresses one of the issues uncovered in #11334.

Test Plan: ./validate

Reviewers: simonpj, hvr, austin, goldfire, bgamari

Reviewed By: bgamari

Subscribers: thomie

Differential Revision: https://phabricator.haskell.org/D1757

GHC Trac Issues: #11334
}}}

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11334#comment:6>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: bgamari Type: bug | Status: new Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): This fails, {{{#!hs {-# LANGUAGE DataKinds #-} import Data.Typeable import Data.Functor.Compose main :: IO () main = print $ typeOf (undefined :: Proxy 'Compose) }}} with a compiler panic due to `TcInteract.mk_typeable_pred` calling `typeKind` on an ill-kinded type, `TYPE 'Lifted (TYPE 'Lifted *) -> Compose * * *`. Note the application `TYPE 'Lifted (TYPE 'Lifted *)`, which oversaturates `TYPE`. The problem here is apparently that the kind variables of `Compose` are all instantiated with `*` with `-XNoPolyKinds` I haven't yet determined where this occurs. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:8 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: bgamari Type: bug | Status: new Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Comment (by goldfire): Oh dear. Something seems terribly terribly wrong. {{{#!hs newtype Compose (f :: k -> *) (g :: k1 -> k) (a :: k1) = Compose {getCompose :: f (g a)} }}} This gives `Compose :: forall k k1. (k -> *) -> (k1 -> k) -> k1 -> *`. But your code above has '''three''' `*`s passed to `Compose`. Somehow the third one is wrong. (I think the first two are correct, as set in `quantifyTyVars`. I'm unconvinced that `quantifyTyVars` is to blame.) GHC is getting duped into thinking `* :: * -> *`, which would, actually, make the `TYPE 'Lifted (TYPE 'Lifted *)` bit well kinded. I've no clue where this is going wrong, but it's going wrong rather spectacularly. My next step would be `-ddump-tc-trace -fprint-explicit- kinds` and search for the first occurrence of `Compose * * *`. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:10 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: bgamari Type: bug | Status: new Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Comment (by simonpj): Easier example is good (comment:11). Hint: always debug a compiler built with `-DDEBUG`. My build shows {{{ decideKindGeneralisationPlan type: (Proxy k1_aCf[tau:5] ('Other k_aCn[tau:5] f_aCo[tau:5] a_aCp[tau:5] |> Other k_aCn[tau:5] f_aCo[tau:5] a_aCp[tau:5]>_N) |> <*>_N) ftvs: [k1_aCf[tau:5], f_aCo[tau:5], a_aCp[tau:5], k_aCn[tau:5]] should gen? True writeMetaTyVar k_aCn[tau:5] := * writeMetaTyVar f_aCo[tau:5] := * WARNING: file compiler\typecheck\TcMType.hs, line 553 Ill-kinded update to meta tyvar f_aCo[tau:5] :: k_aCn[tau:5] -> * * -> * := * :: * * writeMetaTyVar a_aCp[tau:5] := * quantifyTyVars globals: [] nondep: [] dep: [k_aCn[tau:5], f_aCo[tau:5], a_aCp[tau:5]] dep2: [] quant_vars: [] }}} Note the `WARNING`. The bug is this code in `quantifyTyVars`: {{{ -- In the non-PolyKinds case, default the kind variables -- to *, and zonk the tyvars as usual. Notice that this -- may make quantifyTyVars return a shorter list -- than it was passed, but that's ok ; poly_kinds <- xoptM LangExt.PolyKinds ; dep_vars2 <- if poly_kinds then return dep_kvs else do { let (meta_kvs, skolem_kvs) = partition is_meta dep_kvs is_meta kv = isTcTyVar kv && isMetaTyVar kv ; mapM_ defaultKindVar meta_kvs ; return skolem_kvs } -- should be empty }}} But in this case the variables we are quantifying over (I hardly know whether to call them type or kind variables now) are: {{{ k_aCn :: * f_aCo :: k_aCn -> * a_aCp :: k_aCn }}} These appear in the user-written type signature which elaborates to: {{{ Proxy ... (Other k_aCn f_aCo a_aCp) }}} It's clearly NOT RIGHT in the non-poly-kind case to default the "kind" variables to `*`. I guess we have to use `Any` in the same way that we do for types. Can't do more tonight. Richard how are you set? -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:12 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: bgamari Type: bug | Status: new Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Comment (by goldfire): Yech. First off, let's rename the data constructor and type constructor differently. I got quite confused about that! {{{ data Ty (f :: k -> *) (a :: k) = Con (f a) }}} This produces {{{ Ty :: forall k. (k -> *) -> k -> * Con :: forall k (f :: k -> *) (a :: k). (f a) -> Ty k f a }}} The question is: how should we interpret `Proxy 'Con` with `-XNoPolyKinds`? The old rule for `-XNoPolyKinds` was "set all kind variables to `*`", which is still what `quantifyTyVars` is doing. This used to make sense, because all kind variables used to have sort `BOX`, never something like `BOX -> BOX`. But those halcyon days are now gone. I suppose it would be reasonable to set all kind variables of kind `*` to be `*`, and use `Any` for the others. We don't want to just use `Any` all the time, because that wouldn't be backwards compatible and defaulting to `*` is quite sensible. The other reasonable way forward is to issue an error at `Proxy 'Con` without having more direction. For example, the user could say `Proxy ('Con :: Maybe Bool -> Ty * Maybe Bool)` to get the right instantiation. I actually prefer the "issue an error" option. The user could use a kind signature, but more likely should just enable `-XPolyKinds`. Using a promoted data constructor taken from a polykinded datatype without `-XPolyKinds` is asking for trouble. I'm happy to put this change in when I get to rounding up these tickets. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:14 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: bgamari Type: bug | Status: new Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Comment (by goldfire): `Proxy :: Proxy 'Proxy` should also be an error in my "issue an error" option. It's a use of a promoted data constructor of a datatype whose type parameters are not all `*`. To be honest, `-XNoPolyKinds` always confuses me now. :) -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:16 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: goldfire Type: bug | Status: new Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Changes (by simonpj): * owner: bgamari => goldfire Comment:

I suppose it would be reasonable to set all kind variables of kind * to be *, and use Any for the others. We don't want to just use Any all the time, because that wouldn't be backwards compatible and defaulting to * is quite sensible. The other reasonable way forward is to issue an error at `Proxy 'Con` without having more direction

I think I favour: * Default kind vars of kind `*` to `*` * Don't default others; instead error. But note that if we have `{k1:k2, k2:*}` then defaulting `k2` to `*` might mean we could then default `k1`. We do need to do deafulting somehow because even without `PolyKinds` consider {{{ data T f a = MkT (f a) }}} We get `{f :: k -> *, a :: k}` and we can't reject the program. Anyway, Richard, thanks for saying you'll get to it. Simon -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:18 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: goldfire Type: bug | Status: patch Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: Type of failure: Compile-time | Unknown/Multiple crash | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Comment (by Richard Eisenberg ): In [changeset:"84c773e1bc5c551b0f922c6fe9c70762d184a394/ghc" 84c773e/ghc]: {{{ #!CommitTicketReference repository="ghc" revision="84c773e1bc5c551b0f922c6fe9c70762d184a394" Fix #11334. Now we fail when trying to default non-*-kinded kind variables with -XNoPolyKinds. test case: dependent/should_fail/T11334 [skip ci] }}} -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:20 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler

#11334: GHC panic when calling typeOf on a promoted data constructor -------------------------------------+------------------------------------- Reporter: RyanGlScott | Owner: goldfire Type: bug | Status: closed Priority: highest | Milestone: 8.0.1 Component: Compiler (Type | Version: 8.0.1-rc1 checker) | Resolution: fixed | Keywords: TypeInType Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: Compile-time | Test Case: crash | dependent/should_fail/T11334 Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Phab:D1757 Wiki Page: | -------------------------------------+------------------------------------- Changes (by bgamari): * status: merge => closed * resolution: => fixed Comment: Merged as 44a95c6f2421c265da6d20af8c2968933b0d1878. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/11334#comment:22 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler